Midjourney just launched an experimental "Thinking Mode" on its Alpha website to tackle prompt accuracy, typography, and image coherence. Users can trigger it via a "Rerun (Thinking)" button in the job lightbox. Why it matters: reasoning-style processing is finally hitting AI image generation, promising far fewer butchered letters and missed instructions.
Three AI safety researchers fired by OpenAI just fired back with an open letter, denying company claims that they mishandled confidential data. They warn that sudden terminations over standard external collaborations are creating a chilling effect across the company's safety culture. OpenAI maintains the firings followed clear policy violations, not retaliation for raising safety concerns.
OpenAI just launched GPT-6.1 Sol Ultrafast alongside a revamped instant steering feature. The new model runs up to 8x faster than standard Sol with near-Astra intelligence, while instant steering lets you course-correct outputs in real time without wasting compute. It is rolling out across the API, Codex, and ChatGPT Work.
Anthropic just integrated Claude Design across all Claude chats and Claude Code. You can now type /design directly in the terminal to generate on-brand UI layouts pulled straight from your codebase, fine-tune spacing and color on the canvas, and immediately hand the mockup off to code or tools like Vercel.
OpenAI just launched an Ultrafast mode for its GPT-6.1 Sol model, running up to 8x faster than standard speeds while keeping near-Astra intelligence. Designed for live experiences, outage triage, and app-navigating agents, it hits the API at $12/$60 per million tokens and rolls into Codex and ChatGPT Work for eligible high-tier accounts.
Meta just introduced GenIA, a framework that turns standard photos or video clips into complete 3D objects. Built atop SAM 3D, it continuously checks generated shapes against source images to lock down accurate appearance, position, and scale. Why it matters: It bridges generative 3D models with real-world fidelity without retraining the base model, and the code is open for non-commercial local use.
Meta just published a detailed breakdown countering growing backlash over AI data centers. The company claims its liquid-cooled AI facilities consume less water annually than an average U.S. golf course, and that it funds its own grid upgrades to keep local utility bills from rising. Why it matters: Big tech is fiercely defending its infrastructure footprint as public scrutiny around AI's environmental toll heats up.
Epic Games just launched Unreal MCP, an official tool connecting agentic LLMs directly to Unreal Engine and UEFN. Built on the open MCP standard, it allows AI agents to trigger actions and run workflows right inside the engine editor. It means developers can now hand off tedious editor tasks to coding agents.
Google just released Google AI Edge Foresight, a free experimental note-taking app that transcribes meetings entirely offline on Apple Silicon Macs. Powered by its on-device EmbeddingGemma 2 model, the tool turns shorthand bullet points into polished notes and lets you query your meeting audio and local files. Best of all, none of your private data ever leaves your computer.
PiChat just linked LLMs and persistent execution with iPhone Shortcuts to automatically trigger tasks right from incoming notifications. Powered by PiDurableKit, it writes dialogue and tool calls to storage before running, meaning it can resume without losing progress even if iOS kills the app in the background.
HumanX just open-sourced a full-stack framework that turns standard human videos into agile interaction skills for humanoid robots. Tested on the Unitree G1 in simulation, it reproduces basketball pump-fakes, shots, and soccer kicks from monocular footage without motion-capture suits. It basically gives developers a way to train physical robot policies directly from everyday video.
Shenchen Kaidong just built an autonomous robot practice partner called XA1 that predicts shuttle trajectories and repositions itself to return shots without any teleoperation. It handles clears, net drops, and smashes with an over 96% hit rate, even setting a Guinness World Record with 1,452 consecutive counter-hits. Why it matters: badminton is a brutal benchmark for embodied AI, proving robots can coordinate vision and fast mobility in real-time.
Developer wheresryan22 just launched Anatomy, an agent skill for Claude Code and coding agents. Give it an abstract idea like rate limiting, and it builds an interactive isometric machine with moving SVG parts and WebGL shaders instead of boring arrow diagrams.
Anthropic just overhauled its Claude Usage Policy as its models take on more independent real-world work. Effective November 12, the update explicitly outlaws influence networks, weapon guidance systems, and arming drones. Most intriguing is a new rule prohibiting sustained, needless cruelty toward models, backing Claude's ability to hang up on persistently abusive users.
Anthropic just launched Claude Dashboards and Motion in beta, letting users turn raw enterprise data into auto-refreshing charts and generate code-powered animated explainer videos exported as MP4s. Dashboards connects directly to platforms like Snowflake and BigQuery on paid plans, while Motion is available for Team and Enterprise tiers. Meanwhile, Claude Docs, Slides, and Design are now officially out of beta and free for everyone.
Astrophysicist Brice Ménard just used Claude Science to construct the first complete ultraviolet map of the entire sky. Claude orchestrated teams of AI agents to calibrate tens of thousands of satellite observations and inpaint the missing one-third of the sky, predicting unseen UV data within 10% of real measurements. It solves a decades-old observational gap while showing how AI agents can quietly clear researchers' backlogs.
Anthropic just launched the Anthropic Cyber Mission, deploying Claude models and threat researchers to protect critical infrastructure alongside partners like CrowdStrike and Palo Alto Networks. The initiative also debuts OSS Scanner, a free tool offering periodic vulnerability scans and patch recommendations for open-source repositories with a targeted 90% true-positive rate. Why it matters: Attackers already use cutting-edge AI to find exploits, and Anthropic is trying to shift the advantage back to defenders before public infrastructure gets hit.
OpenAI just banned two sophisticated influence operations from Russia and Iran that used ChatGPT to run false-front campaigns. The threat actors built fake journalist personas and an unwitting local think tank to plant geopolitical propaganda directly into mainstream publications. It marks the first Category 5 breach disrupted by the AI maker.
Google just launched the Gemini agent, an enterprise-wide autonomous assistant embedded directly inside Gmail, Docs, Sheets, and Slides. Instead of giving step-by-step instructions, users can assign an end goal and let the agent orchestrate sub-agents, manage persistent memory, and run cross-system workflows in the background. It turns prompt boxes into automated digital coworkers.
Anthropic just committed $150 million over three years to the Genesis Mission, bringing Claude and Claude Code to more than 15 federal science agencies, including NASA and the National Science Foundation. The initiative will grant API credits, tools, and technical support to several hundred research projects focusing on fusion energy and quantum computing.
A solo developer just open-sourced Tonefold, a free desktop composer for macOS and Windows that turns text prompts into multi-track music. Instead of generating flat audio files, it writes 11 instrument layers directly into an editable piano roll using Claude or local Ollama models. Why it matters: musicians get full control over every note, velocity, and MIDI export without dealing with black-box audio generators.
A developer just released seven open-source alternatives to Adobe's flagship apps, using Anthropic's Claude Opus 5.5 to write clean-room clones in Rust. Dubbed Artcraft, the suite mimics Photoshop, Premiere, and Lightroom with zero Adobe code under MIT/Apache licenses. While complex PSD files and plugins are still half-baked, it shows how AI is dismantling the moat around expensive legacy software.
Grok Bot just rolled out full search, reading, and monitoring features across X to all users, with zero connector setup required. You can now tell the bot to track product feedback, watch breaking stories, or build weekly roundups directly in chat. Basically, real-time social listening just became a native feature for anyone on the platform.
Microsoft just unveiled the Surface Laptop Ultra starting at $2,600, powered by Nvidia's new RTX Spark chip to run AI models entirely on-device. The company also rolled out a revamped Windows 11 featuring "Execution Containers" to sandbox rogue AI agents, while dangling a $1,000 trade-in discount for MacBook Pro owners to lure devs back from Apple.
Microsoft just updated Copilot Studio so developers can build user-facing apps, deterministic workflows, and reasoning agents all in one unified workspace. The platform now generates interactive app interfaces from plain text prompts and introduces hooks to inspect or block tool calls during an agent's run. It basically stops treating agents as standalone chatbots and turns them into complete enterprise software solutions.
Google just rolled out its SynthID Detector globally in English, letting anyone verify whether images, video, or audio were AI-generated. The free tool supports content from Google and partners including OpenAI, NVIDIA, Kakao, and soon Apple. Why it matters: After watermarking over 180 billion files, Google is moving AI content verification from closed newsrooms straight to the public.
Google just launched Playground, an experimental browser-based platform that lets anyone build, play, and share custom games using plain text prompts. Creators can tweak physics, rules, and art styles via chat, play multiplayer with friends, and share games publicly. It is currently available to users 18+ in the U.S., with Unity Spark integration coming soon.
Liquid AI just released d1-3B, an open-source multimodal decision model built on its LFM2.5 architecture. Instead of generating conversational text, it evaluates states and questions in a single forward pass with zero output tokens, clocking in at 8 ms on an RTX 4090 and 50 ms on a Jetson Orin Nano. It provides lightning-fast triage and agent guardrails directly on edge hardware.
Anthropic just launched Claude Haiku 5.5, its fastest and most affordable small model yet, running roughly 75% cheaper than Haiku 4.5. It introduces adjustable effort settings designed for high-volume tasks and agentic subagent workflows, while Sonnet 5.5 cache read prices were slashed by half. It gives developers high-speed intelligence without eating through API budgets.
OpenAI just launched GPT-6 and a new Intelligent UI feature for all ChatGPT users, replacing plain text answers with streamable, interactive components like forms, charts, and custom tools on the fly. The model can also interleave thinking with answering, starting search responses up to 44% faster. Why it matters: software is shifting from fixed interfaces to on-demand experiences tailored entirely to your prompt.
SpaceX's Grok Bot is ditching the single-model setup to route every task to the best AI available. Elon Musk announced the bot will dynamically call external APIs, using Anthropic's Claude Opus 5.5 for text, Midjourney for images, and Suno for music generation to deliver the highest quality results.
Google just rolled out Nano Banana 2.1 across Gemini, AI Studio, and its Flow platform. The update focuses on tighter subject consistency, mask-based micro-edits, and blending up to 14 reference photos at up to 4K resolution. Why it matters: While early hands-on impressions suggest it still trails top-tier frontier models in raw fidelity, Google is handing out dozens of free daily generations on Flow.
OpenAI just launched Decisions API in public beta for all developers, powered by GPT-6 Luna. It picks models, tools, or actions up to 10x faster than standard Responses API calls using text and image inputs. Basically, it gives your apps a split-second router to steer requests and flag risky tool calls on the fly.
Musubi just released PolicyLM-1.7B, an open-weight content moderation classifier under the Apache 2.0 license. Instead of relying on static categories, it evaluates your custom policies on every message in under 100ms (35ms median on an L4 GPU). It provides instant, retraining-free guardrails for live chat and AI apps.
OpenAI just rolled out a new Meetings plugin in beta for the ChatGPT Mac app, letting Pro and Business subscribers capture notes hands-free. It pulls from your existing work history to organize personalized takeaways into ChatGPT Space and can instantly draft follow-ups or tweak project plans. Basically, third-party note-taking bots have serious competition right on the desktop.
OpenAI just simplified its API rate limit structure, collapsing five paid tiers into three: Build, Launch, and Grow. The new top tier now requires only $500 in total API spend instead of $1,000, and existing accounts migrate automatically. Scaling up throughput just got twice as easy for fast-growing apps.
Hark just launched Hark Pro, an AI personal assistant built to operate your computer and handle digital chores like replying to emails and renewing car registrations. The app displays its live browser navigation in a small window so users can see every action it takes. Basic access is free, with a paid tier for heavy use.
Google just launched EmbeddingGemma 2, an open-source 740M multimodal embedding model built on Gemma 4 under the Apache 2.0 license. It maps text, code, images, video, and audio into a unified space, requiring as little as 567MB RAM on mobile hardware. Why it matters: Developers can now run privacy-first, fully offline multimodal RAG pipelines on-device without cloud calls.
Google just partnered with the WHO to deploy Earth AI tools that spot disease outbreaks early. Testing during the DRC Ebola outbreak, its Geospatial Reasoning agent identified 48 exposed settlements and over 45,500 at-risk people in minutes rather than weeks. Why it matters: Pairing planetary models with mobility data flips public health from scrambling after viruses to cutting them off weeks ahead.
Economist Daron Acemoglu argued that mass automation fears are overblown, predicting AI will replace only 5% of human tasks and add just 1.5% to GDP over the next decade. Writing in MAI's new Humanist Review magazine, he urged developers to focus on augmenting human workers rather than attempting full replacement.
Pinterest just launched Beauty Guides, an AI tool that translates saved hair and nail Pins into ready-to-use salon action plans. It decodes visual inspiration into exact stylist terminology like "balayage" or "root melt," complete with estimated salon times, price ranges, and maintenance advice.
Mistral just introduced Mistral Large 4, a 1-trillion parameter natively multimodal open-weight model with 52 billion active parameters. Nicknamed "le Chonk," it scores 82% on real-world vulnerability reproduction tests and rivals frontier closed systems in visual grounding. Why it matters: it delivers sovereign, high-performance AI for cybersecurity and enterprise workflows, with weights dropping later this month.
OpenAI just released 722 math papers generated by an unreleased internal frontier model, grouped into 372 result families and backed by Lean formal proofs. Each breakthrough took roughly three hours of ChatGPT Pro thinking time on average. To help mathematicians digest and verify the findings, OpenAI is also pledging $5 million to the Simons Foundation.
OpenAI just rolled out its Auto-review feature for free to all signed-in ChatGPT users without touching plan quotas. The mode deploys a secondary agent to watch over the primary model's moves, blocking high-risk or off-track actions automatically. It basically solves prompt approval fatigue on long agent workflows.
Anthropic just expanded its Cyber Verification Program into a three-tier system, relaxing cybersecurity guardrails on Claude Opus 5.5 and Claude Mythos 5.1 for vetted defenders. The update merges Project Glasswing and tailors model capabilities for defensive analysis, red teaming, and critical infrastructure testing. Dual-use cyber models are risky, but Anthropic wants to ensure defenders hold the upper hand.
OpenAI and Atlassian just deepened their enterprise tie-up, plugging GPT-6 Astra and GPT-5.6 directly into the Rovo AI assistant. Powered by Atlassian’s Teamwork Graph, Rovo connects Jira tickets, Confluence docs, and team chats to spot launch blockers. The companies are also building ways for teams to assign tasks directly to AI agents inside Jira.
Researchers just open-sourced MotionSpaceFlow (MSFlow), a lightweight model that turns text prompts into realistic human motions. Weighing around 1.1GB, it runs on consumer GPUs and lets creators pin any joint to a specific coordinate at any frame without task-specific training.
SemiAnalysis just tested monthly AI subscriptions across major labs and found Anthropic's $200 plan delivers roughly 5x more API-equivalent value than OpenAI's equivalent tier on daily-driver models like Opus 5.5 versus GPT-6.1 Sol. The gap widened after OpenAI quietly halved limits on its $200 Pro tier while rolling out a new $500 plan, shifting the calculus for power users.
Instinct AI just added group chat support, letting friends collaboratively plan trips, split tickets, and coordinate carpools in a single thread—even if other people don't use Instinct. It introduces clever permission gates: the group agent can't touch personal data, and adding a new member automatically holds replies until re-approved. Group coordination might finally stop being a logistics headache.
OpenAI just boosted the default inference speed for GPT-6 Astra and GPT-6.1 Sol by around 50%, pushing output from 30 to 50 tokens per second. The rollout applies across all subscription tiers and partner apps like Devin and OpenCode, kicking off a promised 28-day streak of daily shipping.
Reka just unveiled a research preview of Rho-1, a 19B model trained from scratch to handle text, vision, and robotic action tokens in one network. Instead of delegating tasks across specialized pipelines, it keeps an evolving world state inside a single KV cache. Why it matters: it eliminates handoff latency and enables real-time, interactive video simulation.
Google DeepMind just unveiled AlphaProtein Novo, an AI pipeline that builds functional enzymes from scratch without relying on natural templates. The system designs both 3D structures and amino acid sequences, proving itself in wet labs by synthesizing drug compounds and breaking down toxic plastics. The code is now open on GitHub.
Reflection AI just introduced Beam, an open-weight 501B MoE model with 23B active parameters tailored for coding and agentic workflows. Trained across 10.5K Nvidia GB300 GPUs, it matches frontier open models like GLM 5.2 on key benchmarks while using 3 to 4 times less inference compute. Weights drop later this month.
TikTok just rolled out a batch of AI commerce tools, headlined by a conversational Shopping Assistant agent and a one-click Buy Direct checkout. The platform also opened up its Model Context Protocol (MCP) connector to Claude, Perplexity, and Snowflake, allowing advertisers to run campaign workflows through external AI agents.
OpenAI just launched textGrain, an invisible statistical watermarking system for generated text, in response to the EU AI Act. Global API users can opt in now, while ChatGPT and Codex in the EU will turn it on by default in the coming weeks. Why it matters: text detection is notoriously fragile—swapping 25% of words cuts detection to 17%—which is why OpenAI is limiting detector access to researchers for now.
OpenAI just launched a visual ad format inside ChatGPT, rolling out testing later this month to Free and Go users in the US during image generation. The company is pairing the ad units with major attribution integrations like AppsFlyer and LiveRamp, emphasizing that sponsored placements remain separate from generated images and won't influence chatbot answers.
OpenAI’s Codex team just kicked off an intense 28-day sprint, promising to ship one clear, user-facing improvement every single day for coders and work users—or push a full reset if they miss. The push focuses strictly on simplifying workflows, boosting daily efficiency, and landing new models. Why it matters: OpenAI is answering loud user feedback to strip away clutter and make its flagship coding tool much simpler to use.
Vals AI deployed a team of Claude Opus 5.5 agents to discover two promising room-temperature antiferromagnetic semiconductor candidates for next-gen computer memory. The materials produce zero stray magnetic fields while still sorting electron spins, potentially unlocking denser and vastly faster storage chips. However, the findings currently rely on quantum simulations and await physical lab validation.
Researchers from Fudan University, Tencent Hunyuan, and Zhejiang University just open-sourced Prism, a framework designed for native 2K joint video-audio generation. Instead of wasting compute across every pixel, Prism uses dynamic sparse attention to focus on sound-producing and rapidly changing visual regions, speeding up training by 2.5x while keeping audio and video tightly synchronized. Preview weights and inference scripts are already available on Hugging Face.
Donald Trump just announced the creation of the Super Intelligence Force (SIF) to coordinate federal efforts and maintain U.S. leadership in frontier technology. Under a new executive order, U.S. officials are directed to ditch the term "artificial intelligence" in favor of "super intelligence," alongside a non-binding safety pledge signed by top tech CEOs.
ByteDance introduced DMAD, an adversarial distillation method for rapid visual generation. By training a student model to match real data distributions using a shared-backbone discriminator, it compresses 20-to-30-step video diffusion models like MiniMax-H3 down to four denoising steps while preserving fidelity.
Google is clamping down on free Gemini access, restricting non-paying users primarily to its Flash-Lite model starting October 9. Usage limits are also shifting to a compute-based cap refreshing every five hours, while advanced reasoning like Deep Think and the full Pro model are being pushed behind a $19.99 monthly paywall.
YouTube creators with over 300M combined subscribers just launched the Team Human campaign alongside the Center for AI Safety, urging governments to pause the race for smarter-than-human AI. The coalition is collecting public signatures to deliver to lawmakers in Washington and Brussels this fall, demanding international treaties before superintelligence outpaces human control.
DistScene just introduced a framework that turns a single image into a complete 3D scene with both environment and separate objects. Instead of generating isolated assets and stitching them together, it models everything in a shared coordinate system before refining individual geometries. The team plans to open-source the model and dataset.
OpenAI safety lead David Robinson just quit after 3.5 years, publishing an essay in The Atlantic that calls the company's culture "broken." Robinson, who drafted OpenAI's Preparedness Framework, warned that relentless shipping pressure makes real safety review impossible, pointing to rogue agents and unchecked training runs. Why it matters: Another insider departure confirms growing industry fears that frontier labs are sidelining rigorous safety in favor of product speed.
Social media just blew up over claims that Claude Fable 5.5 beats GPT-6.1 Astra, but the matchup is entirely theoretical. Anthropic has zero record or documentation for Fable 5.5, while OpenAI reportedly shelved Astra over security red lines. The supposed benchmark win is basically an extrapolation of existing tiers rather than a leaked breakthrough.
Meta just open-sourced its Muse AI SDK, letting developers build custom AI agent hardware using off-the-shelf parts like Raspberry Pi and ESP32 boards. Suggested DIY builds include color E Ink reminder displays and portable touchscreens. Meta is also giving away 5,000 units of its pre-built Muse Home Link smart controller to jumpstart home automation use cases.
Runway just debuted Praxis-1, an open-weight world action model that adapts to new robots with light fine-tuning, alongside an end-to-end campaign agent called Runway Ads. Co-CEO Cristóbal Valenzuela warned that the US is still obsessed with LLMs while China pours token usage into video models, predicting live physical AI deployments are 12 to 18 months away.
NVIDIA just revealed the 64GB DGX Spark for $4,999, launching October 23 through manufacturer partners like Dell and HP. Powered by the GB10 Grace Blackwell chip, a single unit runs 100-billion-parameter models locally, and two units can cluster directly via cable to pool 128GB memory with up to 1.7x performance. Why it matters: It gives developers a powerful, private sandbox for agent workloads without racking up cloud bills.
Amazon just launched its "Built Together" initiative alongside a formal Data Center Commitment to push back against community backlash over infrastructure expansion. The company addressed common criticisms, noting its average site uses under 13,000 gallons of water daily and generators run only around 10 hours a year. Why it matters: With over 100 data center moratoriums under review across the U.S., Amazon is fighting local pushback to keep its AI infrastructure buildout moving.
Cactus Compute just released Whistle, a tiny speech recognition model that fits into a single 16.9 MB file. It runs entirely on the CPU with zero dependencies, delivers transcription and word-level timestamps across seven languages, and beats the 145 MB Whisper base model on several standard benchmarks.
Midjourney just updated its Alpha site, adding live style previews directly in the sidebar so you can see aesthetics before burning generation credits. The update also lets users assign default parameters to specific folders, groups multiple style references into a neat thumbnail pill, and eliminates crashes for heavy users with over 100K saved images.
Suno just launched Speech into open beta, an audio model that generates spoken voice and original background music together in a single track. Users simply type a poem, idea, or script, describe the voice and musical backdrop, and get back a finished audio piece. Why it matters: it collapses voice generation and scoring into a single prompt.
OpenAI just rolled out a global virtual try-on feature and a product favorites tool inside ChatGPT. Powered by its new ChatGPT Images 2.5 model, users can click a "Try On" button on shopping listings and upload a selfie to see how outfits look. Why it matters: OpenAI is pushing deeper into fashion discovery, taking direct aim at Google and Pinterest.
Black Forest Labs just launched FLUX 3 Image, introducing bounding-box controls for granular composition and text layout. The model supports up to 10 image references, native 4K rendering, and isolated local edits that let you tweak specific spots repeatedly without degrading the rest of the canvas. It is available via their Playground and API.
a16z just released its State of Markets II report, revealing that a whopping 98% of US households still don't spend a dime on AI services. While awareness of tools like ChatGPT is near-universal, consumer monetization remains in its absolute infancy. Why it matters: The gap between casual adoption and opening wallets is massive, leaving an enormous untapped runway as agentic tools mature.
Developers just open-sourced Mosaic Chunk, a memory system that helps AI video generators track objects and scenes long after they leave the frame. By selectively storing key features instead of every single frame, it maintains visual consistency across cuts. The release includes local-ready models based on MiniMax H3 and Ling World Infinity.
Google just rolled out Guided Vision in Gemini Live on Android, turning the AI assistant into a real-time conversational visual partner for blind and low-vision users. Built with accessibility org Aira, it lets users share their camera to hear surroundings described, read small text, and locate dropped items—with Gemini even offering verbal reframing cues if your angle is off. Multimodal AI is finally finding its most meaningful real-world utility.
Microsoft just launched MAI-Voice-2.1 and MAI-Voice-2.1-Flash, two text-to-speech models capable of instant zero-shot voice cloning. The Flash variant drops inference latency to just 45ms at $15 per million characters, designed to eliminate awkward processing pauses in real-time customer support agents across 23 languages.
Shopify just rolled out Canvas, a visual workspace powered by its AI agent, Sidekick. Instead of wrestling with theme code or static panels, merchants can pan across all pages at once, zoom into details, and chat with Sidekick to edit live code in real time. Sidekick even reviews screenshots of its own work to refine layouts. Why it matters: It turns bespoke storefront design from a multi-week coding slog into an interactive creative session.
Tavus just unveiled Griffin, a full-duplex human interaction model that holds real-time video calls with lifelike expressions and gestures. Unlike turn-based bots, it watches and listens simultaneously, nodding mid-sentence or reacting to pauses, tricking 48% of live study participants into thinking it was human. Griffin-Lite is currently available to early testers.
Anthropic just launched Claude Code Mods, letting developers hack the AI coding tool with simple TypeScript functions. Mods can intercept tool calls, rewrite prompts, draw custom UI elements, and even replace native features like /diff. Better yet, you can ask Claude to write and hot-reload mods for itself. Keep in mind that mods run without sandboxing and hold full machine access.
Strands just open-sourced Decider 2B, a small model built strictly for fast decision-making rather than chatting. Built on a Qwen3.5-2B backbone with a custom pointer head, it picks between predefined options with confidence scores in a median 115ms on an RTX 3090. Why it matters: It gives local agents cheap, reliable guardrails before firing expensive LLM calls.
OpenAI just parted ways with three researchers from its safety team for allegedly sharing confidential company information with a third-party AI safety organization. The company stated an internal investigation confirmed sensitive policy violations, arriving just days after reports exposed leadership brushing aside internal safety warnings. It highlights deepening friction between secretive labs and safety researchers seeking external scrutiny.
Google just announced Gemini 4 Argon, a frontier model built for long-horizon reasoning that pushes the output limit to an unprecedented 1 million tokens. Inside Google, it is already refactoring massive C++ codebases and hunting vulnerabilities. Why it matters: giving models the headroom to think across extended trajectories moves AI from an assistant to an autonomous engineer.
Runway just launched Runway Ads, an autonomous engine that generates video and image creatives, publishes them directly to Meta, Google, and TikTok, and uses incoming performance data to generate the next batch. In internal testing, Runway scaled its weekly ad volume from 77 to 900 while cutting cost per subscriber by 41%.
Industry leaders just gathered at the White House to sign the Accord on Super Intelligence, agreeing that companies building advanced AI bear primary responsibility for safe development and deployment. The pact reinforces that commitment through robust internal controls, independent external evaluations, and board-level oversight, with signatories acknowledging accountability when they fall short.
Ideogram just launched Ideogram 4.5, an image editing model designed to eliminate pixel drift and artifact buildup across repeated edits. It supports native 2K and high-res edits up to 24.2 megapixels, letting users tweak cropped details and stitch them back seamlessly. The model is live now on Ideogram's app, MCP, and API starting at $0.008 per edit.
Instagram just added an AI-powered creative assistant to its standalone Edits app to guide creators on how to beat the algorithm. By pulling historical account data like retention, views, and shares, the bot explains why specific videos flopped and suggests trending tracks, hooks, and scripts. It's free up to a cap, with heavier use locked behind Meta One.
